Source-linked AI summary
Behavior-Mining, Generative Conversations, and Collaborative Advisory: the Future of Travel and Tourism Recommender Systems
Alejandro Bellogín, Linus W. Dietz, Francesco Ricci, Pablo Sánchez
TL;DR
TTRSs remain limited in practical adoption amid outdated data, context-sensitive travel choices, and evaluation methods that emphasize prediction over decision support. The paper synthesizes these pitfalls with evidence about travelers’ GenAI use to argue for transparent, conversational, multistakeholder advisors. Travelers use GenAI for travel planning, though only a minority fully trust its answers.
Problem
TTRSs have limited practical adoption amid outdated behavior data, context-dependent choices, and evaluation metrics that do not measure support for travelers’ decision-making.
Source: §2 · p. 5, §2.1 · p. 6, §2.2 · p. 7Method
The paper combines analysis of TTRS research pitfalls with evidence about travelers’ GenAI use to formulate guidance for future systems.
Source: §3.2 · p. 10, §5 · p. 13Results
Travelers are adopting GenAI quickly and find it useful for planning and during trips, although only a minority fully trust its answers.
Source: §5 · p. 13Takeaways & Limitations
TTRS research should pursue transparent conversational advisors, human-AI collaboration, multistakeholder design, and user studies with better evaluation protocols.
Source: §5 · p. 13TTRS user studies generally involve 100–200 mostly Western, educated, democratic, and rich participants, run briefly, and use hypothetical choices rather than actual travel or spending.
Source: §2.1 · p. 6
Abstract
from arXiv · showhide
Since the early adoption of e-commerce, travel and tourism has been a lab for the design of recommender systems: tools that help travelers choose destinations, flights, accommodations, and combine them into itineraries. Data-driven recommendation techniques, ranging from case-based reasoning to reinforcement learning, have been adapted to travelers' needs. The research community has produced multifaceted prototypes of travel and tourism recommender systems (TTRSs), which are context-dependent, multistakeholder-oriented, and more recently, addressing sustainability issues, such as overtourism. Despite this enduring work, TTRSs are not widespread yet. We argue that three limitations can explain this: outdated and sparse data sets used to train and validate TTRSs, algorithms that prioritize prediction accuracy over domain-specific dimensions such as novelty and contextual relevance, and a failure to address the specific needs of travelers. Targeted incremental research could address these limitations, but a disruptive factor has meanwhile entered the ecosystem of tourism information and commercialization platforms: generative artificial intelligence. According to market research, GenAI applications are becoming the primary entry point for travelers planning their trips. This forces research to rethink how TTRSs should be designed and which core techniques should be integrated. We claim that future TTRSs, in addition to offering personalized information filtering, should become more flexible advisors that support decision making, integrating multiple data types and AI techniques, from data mining to natural language processing. Moreover, they must transparently balance the conflicting goals of travelers, service suppliers, platform owners, and local communities. We then outline research targets for building more effective TTRSs, fruitfully combining old and new recommendation techniques.
1. Travel and Tourism Recommender Systems
TTRSs support diverse travel decisions across trip stages, but face context-sensitive, experiential, and multistakeholder challenges alongside sparse, outdated data. The paper argues for combining established recommendation techniques with GenAI to build more flexible solutions.
1.1. Recommendation tasks
- TTRSs recommend destinations, services, POIs, activities, and complete itineraries before, during, and after travel.They can refine pre-trip suggestions using context gathered in-trip and use trip information to improve future recommendations.Source: §1.1 · pp. 1–2
1.2. Recommendation Techniques
- TTRSs personalize recommendations by traveler or group, use contextual factors, and draw on techniques from case-based reasoning to reinforcement learning.Because historical interactions are often insufficient for new users, practical systems commonly use content-based methods, clustering, and implicit feedback.Source: §1.2 · p. 2
1.3. Application Domain Related Challenges
- Travel recommendations must serve heterogeneous travelers whose choices depend on context, while accounting for experiential item qualities and conflicting stakeholder interests.Travel intent varies with season, mobility constraints, social trends, and group activity; tourism areas may also face overtourism.Source: §1.3 · pp. 3–4
1.4. Pitfalls of Travel RSs
- Sparse, outdated interaction data and context-dependent choices make reliable personalization difficult, leaving recommendations often popular, non-personal, and lacking diversity.These pitfalls accompany limited adoption of TTRSs in operational travel platforms.Source: §1.4 · p. 4
1.5. Building Radically Novel Solutions
- GenAI is changing travel search and planning, prompting the paper to propose integrating traditional data-driven recommendation with large language models for solutions aligned with stakeholder needs.The paper reviews deployed GenAI tools and traveler opinions, then combines those findings with TTRS pitfalls to guide future research.Source: §1.5 · pp. 4–5
2. Pitfalls in Travel and Tourism Recommender Research
Despite an established, active TTRS research community, commercial use has remained limited. The section identifies shortcomings in data, algorithms, and evaluation as key pitfalls.
2. Pitfalls in Travel and Tourism Recommender Research
- TTRSs remain uncommon in practice despite substantial research activity, with major platforms previously offering mainly information filtering and generic inspiration.The section groups the technological deficiencies under data sets, recommendation algorithms, and evaluation methods.Source: §2 · p. 5
2.1. Data Sets Pitfalls
- Public tourism behavior data are often over a decade old, sparse, biased, and based on social-network activity rather than direct recommender-system interactions.Check-ins may be timestamped after visits, limiting reliable context and visit sequencing; available records may also reflect implicit preferences only.Source: §2.1 · p. 6
- LBSN data often overrepresent local behavior and omit traveler context, while short, small, hypothetical user studies cannot capture actual travel spending or long-term behavior.User studies commonly involve 100–200 participants and may not measure seasonal behavior, evolving preferences, or repeat visits.Source: §2.1 · p. 6
2.2. Algorithms Pitfalls
- Popularity and exposure bias can reinforce already-visited POIs, while many systems optimize traveler interests and platform revenue rather than the conflicting goals of other stakeholders.In tourism, recommending popular venues may add little to what tourists can find in travel guides; local communities and DMOs may prioritize sustainability and overtourism.Source: §2.2 · pp. 6–7
- Trip context and intentions can change dynamically, but many algorithms are rigid during in-trip recommendations; explanations are also needed to help travelers assess costly POI suggestions.Context may include mood, weather, current intent, group interests, trip phase, and budget; conversational systems could allow real-time interest updates.Source: §2.2 · p. 7
2.3. Evaluation Methodology Pitfalls
- Evaluation also lacks consistent protocols and contextual user-group analysis, while offline metrics omit usability, accessibility, usefulness, and interaction complexity.Global performance can improve for specific users, venues, or periods while worsening the experience for others; inconsistent procedures hinder comparison and reproducibility.Source: §2.3 · pp. 7–8
- Offline evaluations compare recommendations with past visits, but this prediction-focused protocol does not assess discovery, novelty, diversity, catalog coverage, or support for multiple stakeholders.The underlying mismatch is between predicting past interests and supporting travelers throughout decision making.Source: §2.3 · p. 7
3. GenAI and Tourism Information Search
GenAI is reshaping travel search and trip planning, with platforms offering conversational, context-aware tools across travel tasks. Surveys report substantial use and perceived helpfulness, alongside persistent trust concerns.
3. GenAI and Tourism Information Search
- GenAI research directions build on established RS concepts, but the paper questions whether treating recommenders as platform components that manage items is sufficient for the new landscape.The paper situates its discussion amid social and AI-mediated travel information, which may be selected, biased, or outdated.Source: §3 · p. 8
3.1. Initial Adoption of GenAI
- Travel platforms are incorporating LLM-powered search, recommendation, and booking features that support context-aware suggestions and iterative trip planning in natural language.GenAI tools are described as intermediaries across the travel life-cycle, making interactions more adaptive, personalized, and automated.Source: §3.1 · p. 8
- Conversational tools support destination and service searches, POI and activity recommendations, and itinerary planning, sometimes combining external information with user constraints.Examples include refining interests and budgets, incorporating ratings or weather, using real-time prices, and adjusting plans for schedules, proximity, or availability.Source: §3.1 · pp. 8–9
3.2. GenAI Tool Usage and Effectiveness
- Corporate experimentation can amount to “AI washing,” while tourism literature reports value alongside trust and privacy concerns and frames GenAI tools as starting points, not final ones, in travel planning.The passage notes that firms may signal AI adoption with little measurable return and at the cost of customer trust.Source: §3.2 · pp. 9–10
- More than half of surveyed U.S. travelers had tried GenAI and one in three had used it for travel; 78% found its results helpful, but only about one-third fully trusted responses.A separate survey found that 89% wanted to use AI in future travel planning, while only 6% fully trusted AI.Source: §3.2 · pp. 9–10
- In Japan, 77.8% of GenAI users had used it for travel; anticipated uses included creating itineraries, suggesting routes, searching and booking transport, and finding local gourmet information.Respondents also expected helpful real-time information about tourist attractions and congestion levels.Source: §3.2 · p. 10
4. The Disruptive Potential of GenAI
The authors argue GenAI should complement rather than replace TTRSs: conversational flexibility joins recommendation-system strengths, and five research goals connect to nine opportunities for addressing TTRS pitfalls.
4. The Disruptive Potential of GenAI
- The proposed direction combines GenAI’s conversational flexibility and content generation with recommender systems’ preference learning, structured-data exploitation, and evaluation methods.The authors envision flexible, dynamic conversational advisors rather than rigid information-filtering tools.Source: §4 · p. 10
4.1. Personalizing the Interaction, Not Only the Content
- Personalized, iterative dialogue can elicit travel context, adapt recommendations to changing intent, and explain suggestions when explanations faithfully reflect selection.The paper identifies travel purpose, group composition, budget, and mood as useful context to elicit.Source: §4.1 · p. 11
4.2. Leveraging the Value of Unstructured Content
- GenAI can process reviews and narratives for experiential preference signals, but recommendation should combine them with ratings, timetables, and contextual data.Continuously written content can reduce reliance on old interaction logs and reveal qualities check-in data cannot capture.Source: §4.2 · p. 12
4.3. Supporting Informed and Sustainable Travel Decisions
- Recommender advisors can flag unreliable claims, steer travelers toward sustainable options, and support fairer group decisions through cooperative planning.Countering popularity bias requires integrating recommender-system debiasing and diversification techniques, since GenAI may amplify the bias.Source: §4.3 · p. 12
4.4. Generating Targeted Information and Promotion Content
- GenAI can help destinations create tailored recommendable items and synthetic interactions, while simulations estimate effects on visitor flows and revenue distribution.Stakeholder-side tools may support strategic traveler distribution and fairer personalized bundles.Source: §4.4 · p. 12
4.5. Helping Travelers to Choose and Enjoy Better Experiences
- Systems should help travelers anticipate satisfaction and persuasive effects, evaluate success through better choices and experiences, and provide context-aware in-trip guidance.The authors also propose personalized instructions and dynamic itinerary replanning at the destination.Source: §4.5 · pp. 12–13
5. Conclusions
The paper calls for human-AI collaboration to make TTRSs transparent, multistakeholder conversational advisors, with research and evaluation that address trust, bias, and real-world effects.
5. Conclusions
- TTRS research should embrace human-AI collaboration and develop more transparent conversational advisors for travelers, destinations, and service providers.The conclusion identifies multistakeholder system design and evaluation as key.Source: §5 · p. 13
- Evaluation should use user studies that measure evolving engagement, satisfaction, and trust, while monitoring GenAI-related biases and persuasive nudges’ real-world effects.The authors also call for simulation studies that better match traveler decision-making and marketing interventions before deployment.Source: §5 · p. 13
- The authors argue that integrating GenAI with recommender-system methodology can address long-standing TTRS pitfalls, rather than setting aside two decades of RS foundations.Source: §5 · p. 13